1.2 Conditional probability and independence

Definition of conditional probability

Conditional probability measures the probability of event A occurring given that event B has already occurred. It's the tool you use to update probabilities when you learn new information, and it shows up constantly in actuarial work: adjusting risk estimates as new data arrives, pricing policies based on policyholder characteristics, and evaluating diagnostic tests.

Formula for conditional probability

The conditional probability of A given B is written as $P(A|B)$ and defined by:

$P(A|B) = \frac{P(A \cap B)}{P(B)}$

where $P(A \cap B)$ is the joint probability of both A and B occurring, and $P(B)$ is the marginal probability of B. This formula requires $P(B) > 0$, since conditioning on a zero-probability event is undefined.

Notation

• $P(A|B)$ reads as "the probability of A given B."
• The vertical bar "|" separates the event of interest (left) from the conditioning event (right).
• You may also see $P_B(A)$, which emphasizes that we're evaluating A's probability within the restricted world where B happened.

Intuitive explanation

Think of conditional probability as shrinking the sample space. Once you know B occurred, you ignore every outcome outside B. You then ask: of the outcomes in B, how many also belong to A?

For a concrete example, consider drawing a card from a standard 52-card deck. What's the probability of drawing a king (event A) given the card is red (event B)? There are 26 red cards and 2 red kings, so:

$P(\text{King}|\text{Red}) = \frac{2}{26} = \frac{1}{13}$

Notice this equals the unconditional probability of drawing a king ($\frac{4}{52} = \frac{1}{13}$), which makes sense because "king" and "red" are independent in a standard deck.

Properties of conditional probability

Conditional probability satisfies the same axioms as ordinary probability: non-negativity, the total probability of the restricted sample space equals 1, and countable additivity holds. This means all the standard probability rules still apply once you condition on an event.

Law of total probability

The law of total probability lets you compute $P(A)$ by breaking the sample space into pieces. If $B_1, B_2, \ldots, B_n$ form a partition of the sample space (mutually exclusive and exhaustive), then:

$P(A) = \sum_{i=1}^{n} P(A|B_i) \cdot P(B_i)$

This is especially useful when you can't compute $P(A)$ directly but you can compute it within each partition element. For instance, an insurer might partition policyholders into risk classes and compute the overall claim probability as a weighted average across classes.

Bayes' theorem

Bayes' theorem lets you "reverse" a conditional probability. It connects $P(A|B)$ to $P(B|A)$:

$P(A|B) = \frac{P(B|A) \cdot P(A)}{P(B)}$

The terms have specific names in Bayesian language:

• $P(A)$ is the prior probability of A (your belief before observing B).
• $P(B|A)$ is the likelihood of observing B if A is true.
• $P(B)$ is the marginal probability of B (often computed via the law of total probability).
• $P(A|B)$ is the posterior probability of A after observing B.

Bayes' theorem is the engine behind updating risk assessments as new claims data, medical test results, or other evidence comes in.

Multiplication rule

The multiplication rule rearranges the conditional probability formula to give you joint probabilities:

$P(A \cap B) = P(A|B) \cdot P(B) = P(B|A) \cdot P(A)$

For more than two events, this generalizes via the chain rule:

$P(A_1 \cap A_2 \cap \cdots \cap A_n) = P(A_1) \cdot P(A_2|A_1) \cdot P(A_3|A_1 \cap A_2) \cdots P(A_n|A_1 \cap \cdots \cap A_{n-1})$

Each factor conditions on everything that came before it. This is particularly useful for sequential processes like drawing cards without replacement or modeling multi-stage claim events.

Independence vs dependence

Independence and dependence describe whether knowing one event occurred changes the probability of another. Getting this distinction right is critical; assuming independence when events are actually dependent (or vice versa) leads to serious errors in premium calculations and risk models.

Definition of independence

Two events A and B are independent if and only if:

$P(A \cap B) = P(A) \cdot P(B)$

Equivalently, $P(A|B) = P(A)$ and $P(B|A) = P(B)$. Learning that B occurred tells you nothing new about A.

If this condition fails, the events are dependent.

Checking for independence

To verify independence:

1. Compute $P(A \cap B)$ (the joint probability).
2. Compute $P(A) \cdot P(B)$ (the product of marginals).
3. If they're equal, the events are independent. If not, they're dependent.

Alternatively, check whether $P(A|B) = P(A)$. If conditioning on B doesn't change A's probability, the events are independent.

Examples of independent events

• Successive fair coin tosses: The probability of heads on the second toss is 0.5 regardless of the first toss outcome. Each toss is independent.
• Rolling a die and drawing a card: The die outcome has no physical connection to the deck, so these events are independent.

Examples of dependent events

• Drawing cards without replacement: If the first card drawn is an ace, only 3 aces remain among 51 cards. The probability of drawing an ace on the second draw drops from $\frac{4}{52}$ to $\frac{3}{51}$.
• Consecutive-day weather: Rain today increases the probability of rain tomorrow because weather systems persist. The events "rain today" and "rain tomorrow" are dependent.

Conditional probability with multiple events

When problems involve three or more events, the same conditional probability framework extends naturally, but you need to be careful about the distinction between pairwise and mutual independence.

Conditional probability for three or more events

The conditional probability of A given both B and C is:

$P(A|B \cap C) = \frac{P(A \cap B \cap C)}{P(B \cap C)}$

This generalizes to any number of conditioning events:

$P(A|B_1 \cap B_2 \cap \cdots \cap B_n) = \frac{P(A \cap B_1 \cap B_2 \cap \cdots \cap B_n)}{P(B_1 \cap B_2 \cap \cdots \cap B_n)}$

The denominator must be positive for the expression to be defined.

Conditional independence

Events A and B are conditionally independent given C if:

$P(A \cap B|C) = P(A|C) \cdot P(B|C)$

Once you know C occurred, learning about A gives you no additional information about B (and vice versa).

A key subtlety: conditional independence given C does not imply unconditional independence, and unconditional independence does not imply conditional independence given C. These are separate properties, and you need to verify each one on its own terms.

Pairwise vs mutual independence

For three events A, B, and C:

• Pairwise independence means every pair is independent: $P(A \cap B) = P(A)P(B)$, $P(A \cap C) = P(A)P(C)$, and $P(B \cap C) = P(B)P(C)$.
• Mutual independence requires pairwise independence plus the three-way condition: $P(A \cap B \cap C) = P(A) \cdot P(B) \cdot P(C)$.

Mutual independence is strictly stronger. There are classic examples where all three pairs are independent but the triple product condition fails. On exams, if a problem says events are "independent," it typically means mutually independent unless stated otherwise.

Applications of conditional probability

Insurance and risk assessment

Insurers use conditional probability to segment policyholders by risk factors. For example, $P(\text{claim}|\text{driver under 25})$ might be significantly higher than $P(\text{claim}|\text{driver over 25})$. This drives risk-based pricing: young drivers pay higher premiums because their conditional claim probability is higher.

The law of total probability lets an insurer compute the overall portfolio claim rate by weighting each risk class's conditional claim probability by the proportion of policyholders in that class.

Medical testing and diagnosis

Medical tests have two key conditional probabilities:

• Sensitivity: $P(\text{positive test}|\text{disease present})$
• Specificity: $P(\text{negative test}|\text{disease absent})$

But what patients and doctors actually want to know is the reverse: $P(\text{disease}|\text{positive test})$. Bayes' theorem bridges this gap. When a disease is rare (low prevalence), even a highly sensitive and specific test can produce a surprisingly low posterior probability of disease given a positive result. This is a classic exam topic and a real-world source of confusion.

Machine learning and classification

• Naive Bayes classifiers assume features are conditionally independent given the class label, then apply Bayes' theorem to predict the most probable class. Despite the "naive" independence assumption, these classifiers work well in practice for tasks like spam filtering.
• Hidden Markov models use conditional probabilities to relate observed data to hidden states over time, with applications in speech recognition, bioinformatics, and financial modeling.

Common misconceptions and pitfalls

Confusing conditional and joint probability

$P(A \cap B)$ and $P(A|B)$ are different quantities. The joint probability asks "what's the chance both happen?" while the conditional asks "given B happened, what's the chance of A?" Plugging one where the other belongs, especially inside Bayes' theorem, will give you a wrong answer. Always check which quantity the problem is asking for.

Assuming independence without verification

Don't assume events are independent just because it seems convenient. Independence is a precise mathematical condition that must be verified or justified. Incorrectly assuming independence between correlated risks (e.g., claims from policyholders in the same geographic area during a natural disaster) can lead to severe underestimation of aggregate risk.

Reversing the conditional (the "prosecutor's fallacy")

Confusing $P(A|B)$ with $P(B|A)$ is one of the most common errors. For example, $P(\text{positive test}|\text{disease})$ (sensitivity) is not the same as $P(\text{disease}|\text{positive test})$ (positive predictive value). These can differ dramatically, especially when the base rate of the disease is low. Always use Bayes' theorem to reverse a conditional.

Solving conditional probability problems

Step 1: Identify events and given information

Read the problem carefully and label the events. Determine which event is the "given" (conditioning event) and which probability you need to find. Note whether you're given joint probabilities, marginals, or conditional probabilities.

Step 2: Organize with a tree or table

• Probability trees work well for sequential problems (e.g., draw a card, then draw another). Each branch represents an outcome with its probability, and you multiply along branches to get joint probabilities.
• Two-way tables work well for problems with two categorical variables. Fill in joint, marginal, and conditional probabilities in a matrix format.

Step 3: Select and apply the right formula

• If you know $P(A \cap B)$ and $P(B)$: Use the definition directly: $P(A|B) = \frac{P(A \cap B)}{P(B)}$
• If you need a joint probability: Use the multiplication rule: $P(A \cap B) = P(A|B) \cdot P(B)$
• If you need to compute a marginal probability from conditional pieces: Use the law of total probability: $P(A) = \sum_{i} P(A|B_i) \cdot P(B_i)$
• If you need to reverse a conditional: Use Bayes' theorem: $P(A|B) = \frac{P(B|A) \cdot P(A)}{P(B)}$

Step 4: Verify your answer

Check that your answer is between 0 and 1. If the problem gives you enough information, verify by computing the same probability a different way (e.g., using a table vs. a formula). Confirm that you haven't accidentally swapped the direction of the conditional.